Aging effects on discrimination learning, logical reasoningand memory in pet dogs

Lisa J. Wallis & Zsófia Virányi & Corsin A. Müller &

Samuel Serisier & Ludwig Huber & Friederike Range

Received: 3 August 2015 /Accepted: 16 December 2015 /Published online: 4 January 2016# The Author(s) 2016. This article is published with open access at Springerlink.com

Abstract In laboratory dogs, aging leads to a decline invarious cognitive domains such as learning, memoryand behavioural flexibility. However, much less isknown about aging in pet dogs, i.e. dogs that are ex-posed to different home environments by their care-givers. We used tasks on a touchscreen apparatus todetect differences in various cognitive functions acrosspet Border Collies aged from 5 months to 13 years.Ninety-five dogs were divided into five age groupsand tested in four tasks: (1) underwater photo versusdrawing discrimination, (2) clip art picture discrimina-tion, (3) inferential reasoning by exclusion and (4) amemory test with a retention interval of 6 months. Thetasks were designed to test three cognitive abilities:visual discrimination learning, logical reasoning andmemory. The total number of sessions to reach criterionand the number of correction trials needed in the two

discrimination tasks were compared across age groups.The results showed that both measures increased linear-ly with age, with dogs aged over 13 years displayingslower learning and reduced flexibility in comparison toyounger dogs. Inferential reasoning ability increasedwith age, but less than 10 % of dogs showed patternsof choice consistent with inference by exclusion. No ageeffect was found in the long-term memory test. In con-clusion, the discrimination learning tests used are suit-able to detect cognitive aging in pet dogs, which canserve as a basis for comparison to help diagnosecognition-related problems and as a tool to assist withthe development of treatments to delay cognitivedecline.

Keywords Touchscreen . Learning . Flexibility .

Reasoning by exclusion . Logical reasoning .Workingmemory. Long termmemory . Dog

The development and aging of cognitive processes suchas learning, memory and logical reasoning and theirinteractions with genetic, environmental and social fac-tors have so far almost exclusively been studied inhumans (Baltes 1987; Craik and Bialystok 2006).Learning and memory are basic processes, which areessential for the acquisition of knowledge, and further-more allow an individual to apply knowledge in novelsituations through logical reasoning. These basic cogni-tive abilities are known to change over the lifespan,increasing rapidly from infancy to young adulthoodand then, depending on the specific ability, are either

AGE (2016) 38: 6DOI 10.1007/s11357-015-9866-x

Electronic supplementary material The online version of thisarticle (doi:10.1007/s11357-015-9866-x) contains supplementarymaterial, which is available to authorized users.

L. J. Wallis (*) : Z. Virányi : C. A. Müller : L. Huber :F. RangeClever Dog Lab, Messerli Research Institute, University ofVeterinary Medicine Vienna, Medical University of Vienna,University of Vienna, Veterinärplatz 1, 1210 Vienna, Austriae-mail: lisa.wallis@vetmeduni.ac.at

L. J. Wallis :C. A. MüllerDepartment of Cognitive Biology, University of Vienna,Vienna, Austria

S. SerisierRoyal Canin Research Center, Aimargues, France

improved (as is the case for knowledge formation),maintained or declined in old age (Baltes 1987; Pearce2008).

Cognitive processes are regulated by executive func-tions comprising selective attention, working memory,flexibility and inhibition, some of which have also beenfound to be particularly sensitive to aging (Cepeda et al.2001; Clark et al. 2006; Manrique and Call 2015; Rapp1990; Tapp et al. 2003a, b; Wallis et al. 2014). There areremarkably few studies in humans or animals whichdetail the changes in these specific cognitive processesand their regulation by executive processes over thecourse of the entire lifespan, as cognitive developmentand aging are frequently disassociated. Previous studiesin humans using cognitive batteries showed that learn-ing and logical reasoning increase rapidly from infancyto young adulthood and then decline steadily (Craik andBialystok 2006; Moshman 2004) and that long-termmemory increases into the fifth and sixth decades of lifeand only shows very gradual decline thereafter(Brickman and Stern 2010).

Learning ability is often measured in human andanimal studies using one specific type of learning calleddiscrimination learning. Discrimination learning proto-cols generally utilise a two-choice procedure, where twostimuli are presented, but only one of them leads to areward. Since the stimuli are presented simultaneously,parallel processing is necessary. The subject is requiredto attend to a target stimulus, while ignoring or avoiding‘distractor’ information (Julesz and Schumer 1981).Selection of the target stimulus results in positive rein-forcement, which causes an increase in the frequency ofthe choice of this stimulus (Mell et al. 2005). Deficits insimultaneous processing of stimuli increase with age inhumans and animals, due to decreases in processingspeed, reduced cognitive resources and an inability toignore distracting information (Baddeley et al. 2001;Costello et al. 2010; Lavie 1995; Snigdha et al. 2012).Age-related impairments in learning are shown by anincrease in the number of trials necessary to reach alearning criterion as well as an increase in perseverativeresponding, which is defined as the repetition of aparticular response, such as selection of a particularstimulus, due to an inability to adapt to external feed-back of right and wrong. Perseverative responding maybe a sign of reduced cognitive flexibility, which isthe ability to adjust thinking or attention in re-sponse to changing goals and/or environmentalstimuli (Scott 1962).

Another form of learning is learning by exclusion, atype of logical reasoning defined as the selection of thecorrect alternative by logically excluding other potentialalternatives (Call 2006). Human children are known tolearn by exclusion, which develops from the age of2 years (Heibeck and Markman 1987; Horst andSamuelson 2008; Spiegel and Halberda 2011). Sincechildren as young as 2 years are able to make simpleinferences by exclusion, this ability likely depends onsimple associative learning mechanisms and thereforecan also be found in animals, based on previous positivefindings (Aust et al. 2008; Call 2006; Herman et al.1984; Kaminski et al. 2004; Kastak and Schusterman2002; Pilley and Reid 2011). For example, Aust et al.(2008) found evidence of reasoning by exclusion in petdogs using a touchscreen procedure. Additionally,Kaminski et al. (2004) found that a Border Collie hadthe ability to acquire the relation between a word and theobject that the word refers to (the referent) and that itcould also infer the referent of new words by exclusionlearning and retain this knowledge over time. However,dogs’ preference for novelty could also explainKaminski et al.’s results (see Kaulfuss and Mills(2008)). The study of Pilley and Reid (2011) on anotherBorder Collie ruled out any influence of novelty prefer-ence, by including baseline novelty preference measure-ments (but see Griebel and Oller (2012) for an alterna-tive conclusion on the dogs’ performance).

Currently, there are no studies in non-human animalsdetailing how the ability to reason by exclusion changeswith age over the lifespan. Studies in humans, however,have demonstrated that logical reasoning ability is close-ly related to an individual’s working memory capacity,which is limited in complex tasks (Kyllonen andChristal 1990; Süß et al. 2002). Working memory ca-pacity can severely limit reasoning abilities particularlyin tasks where time limits are implemented (Chuderski2013). Moreover, in order to reach learning criterions incomplex discriminations and learning by exclusiontasks, long-termmemory is required to store informationsuch as positive and negative stimulus associations indiscrimination learning or the correct labelling of a newword or object in exclusion tasks. While working mem-ory and logical reasoning ability decline with old age(Borella et al. 2008; Brockmole and Logie 2013;De Luca et al. 2003; Lee et al. 2005; Sander et al.2012), long-term memory shows very little declinewhen comparing younger and older adults(Brickman and Stern 2010).

6 Page 2 of 18 AGE (2016) 38: 6

Learning and memory have been extensively studiedin laboratory dogs which are considered to be a goodanimal model for human aging and Alzheimer’s disease,since they develop similar age-related neuropathologiesas humans, as well as a similar decline in their measuresof sensorimotor ability, selective attention, learning,short-term memory and executive function with age(Adams et al. 2000a, b; Head et al. 1995; Head et al.2000; Landsberg et al. 2003; Milgram et al. 1994; Tappet al. 2003a, b; Wallis et al. 2014). For example, likehumans, dogs’ selective visual attention and discrimina-tion learning is sensitive to aging in some tasks(Milgram et al. 2002; Snigdha et al. 2012), whereas inother tasks discrimination learning was not affected byage (egocentric spatial discrimination, Christie et al.2005; object discrimination learning, Milgram et al.1994). This inconsistency in laboratory dogs is likelyexplained by the level of difficulty of the task whichinfluences whether an age effect is detected or not(Adams et al. 2000a, b; Head et al. 1998; Milgramet al. 1994). Previous research has also shown that olderdogs tend to show perseverative responding in complexdiscrimination learning tasks similarly to humans (Grantand Berg 1948; Mell et al. 2005; Tapp et al. 2003b).

Few studies have addressed how long dogs are ableto remember previously learnt discriminations, which isa measure of long-term memory. Araujo et al. (2005)tested laboratory beagles in a working memory task andfound a significant decline with age. In contrast, theirperformance remained stable after a 2-year break periodin previously learned discriminations. Therefore, work-ing memory capacity in dogs declines with age, whereaslong-term memories are more resistant to aging, whichreflects similarities to humans (Adams et al. 2000a, b;Fiset et al. 2003; Fiset 2007; Salvin et al. 2011; Tappet al. 2003b).

Most research projects have relied on laboratory-keptbeagles to examine age-related cognitive changes. Oneadvantage of utilising pet dogs living with human fam-ilies is that we are able to examine the development andaging of cognition under the influence of the humanliving environment. This environment is likely to bemore enriching and stimulating than that found inlaboratory-housed beagles and thus may provide agreater level of resistance to the effects of aging(Milgram et al. 2005). There are few studies which haveexamined age-dependent losses in learning and memoryin companion dogs (González-Martínez Á et al. 2013;Mongillo et al. 2013; Salvin et al. 2011). Such studies

are crucial for the development of objective diagnosticprocedures to enable the accurate diagnosis of caninecognitive dysfunction syndrome (age-related non-normal cognitive decline) and to quantify normal suc-cessful aging in pet dogs outside a laboratory setting.

The use of the touchscreen apparatus allows the designand implementation of non-verbal standardised taskswhich can be utilised to examine cognitive functioningsuch as individual learning abilities, memory and logicalreasoning in non-human animals and permits compari-sons with humans and across species (Spinelli et al. 2004;Steurer et al. 2012). Computerisation results in the elim-ination of social cuing and increases/maintains themotivation to work in the subjects (Range et al.2008). The touchscreen can be used to establish base-line measures of cognitive aging associated with nor-mal aging, which has so far only been utilised inhumans (Clark et al. 2006), laboratory-housed non-human primates (Joly et al. 2014; Nagahara et al.2010) and rodents (Bussey et al. 2008).

Accordingly, the goals of the present study were totest the effect of aging on discrimination learning, rea-soning by exclusion and memory in a cross-sectionalsample of pet dogs ranging in age from 5 months to13 years, in order to determine when dogs cognitivelymature and when cognitive decline begins. After receiv-ing pre-training on how to work on a touchscreen, thedogs were tested in four tasks: (1) underwater photoversus drawing discrimination consisting of six stimuli,(2) clip art picture discrimination consisting of eightstimuli (which were also used as a training for the nexttask on inferential reasoning by exclusion), (3) inferen-tial reasoning by exclusion testing, and (4) a memorytest on the clip art picture discrimination (task 2) per-formed after a 6-month break from the touchscreen. Twodiscrimination tasks were utilised which differed notonly in the types and number of stimuli used but alsoin their difficulty level. In the first discrimination (un-derwater photo vs. drawing), the positive and negativeclass was composed of highly similar members withlarge inter-class and small intra-class differences, where-as the more difficult second discrimination (clip artpictures) had equal inter-class and intra-class differ-ences. Based on previous studies in laboratory dogs,we predicted that dogs’ learning ability will decreasewith age and perseverative responding will increase(Milgram et al. 2002; Snigdha et al. 2012; Tapp et al.2003a). Long-term memory was predicted to remainstable with age (Araujo et al. 2005), and finally, based

AGE (2016) 38: 6 Page 3 of 18 6

on information from the human literature, the ability tomake inferences by exclusion was predicted to peak inyoung adulthood and decline thereafter (Moshman2004), in conjunction with dogs’ working memory abil-ity (Tapp et al. 2003b).

Methods

Subjects

Ninety-five pet dogs ranging in age from 5 months to13 years and 10 months were recruited to participate inthe study (Table 1). All dogs were from one breed,the Border Collie, in order to exclude the effectsof different developmental and aging speeds ofdifferent breeds. The subjects were split into five agegroups according to Siegal and Barlough (1995), whichaimed to reflect the developmental periods in the BorderCollie (late puppyhood, adolescence, early adulthood,middle age and late adulthood which included seniorand geriatric).

Apparatus

Testing was conducted in a room (3×4 m) at the CleverDog Lab in Vienna, Austria. The test apparatusconsisted of a closed rectangular box containing thefood pellet dispenser (feeder box; 48 × 100 × 60 cm(w × h × d) ) and an ad j acen t t e s t i ng n i che(48×100×30 cm) where the touchscreen was locatedalong the top back wall (Fig. 1). Dogs were tested in thetesting niche, which allowed subjects to reach thetouchscreen whilst their vision was shielded to avoidpotential distractions from the side or above, thusminimising human influence on the dogs’ performance.Inside the testing niche, a 15″ TFT 600×800 pixel

resolution computer screen was mounted behind aninfrared touchframe (Carroll Touch, Round Rock, TX,USA; 32 vertical× 42 horizontal resolution (Aust et al.2008; Huber et al. 2005; Range et al. 2008; Steurer et al.2012)). A small hole beneath the touchscreen allowedcommercial dog food pellets to be automatically dis-pensed in order to administer reinforcement for correctchoices. The presentation of the stimuli and the releaseof the reward were controlled by a microcomputerinterfaced through a digital input–output board. Theowner and the experimenter were present during thetesting but were prevented from viewing the stimuli bythe walls of the testing niche (see Fig. 1a for owner andexperimenter locations).

Procedure

The touchscreen training and testing proceduresconsisted of two pre-training steps (an approach trainingand a simple geometric form discrimination) and fourtasks: a ‘categorical’ discrimination (underwater photo-graphs and drawings; task 1), a clip art picture discrim-ination (the training phase of the inferential reasoning byexclusion tests; task 2), inferential reasoning by exclu-sion testing (previously reported in Aust et al. 2008; task3) and finally task 4: a memory test after a 6-monthbreak from the touchscreen consisting of a repetition oftask 2 (clip art picture discrimination/inference by ex-clusion training).

Touchscreen pre-training

Approach training

Dogs visited the lab once a week and participated inthree to four sessions (each session consisted of 30 to 32individual trials), over a half-hour period, with short

Table 1 Age, sex and neuter status of subjects

Age group Life stage Age in months Mean + SD age in years Male (neutered) Female (neutered) Total

Group 1 Late puppyhood 5–12 0.68+ 0.16 7 (0) 13 (1) 20

Group 2 Adolescence >12–24 1.39+ 0.24 10 (1) 12 (2) 22

Group 3 Early adulthood >24–36 2.42+ 0.30 7 (3) 14 (5) 21

Group 4 Middle age >36–72 4.41+ 0.89 5 (2) 13 (6) 18

Group 5 Late adulthood >72 8.61+ 2.10 5 (3) 9 (9) 14

Total 34 (9) 61 (23) 95

6 Page 4 of 18 AGE (2016) 38: 6

breaks in between sessions. Dogs were trained to touchthe monitor with their nose using a clicker-aided shapingprocedure. A stimulus, either a circle or a square, ap-peared in random locations on a black screen. If the dogstouched the stimulus with their nose, the infrared lightgrid was interrupted, which triggered an acoustic signaland delivery of a food treat. After the dog becamefamiliar with the action of touching the stimulus andreceiving the food reward via the automatic feeder(without help from the experimenter), the simple geo-metrical form discrimination was initiated.

Geometric form discrimination

In this task, the subjects were shown a square and acircle side by side. Both stimuli were varied in colourbetween trials (red, yellow or blue, Fig. 2a). The dogswere assigned to two groups balanced for age group andsex. Group ‘square’ was rewarded for touching thesquare; group ‘circle’ was rewarded for touching thecircle. A forced two choice procedure was utilised,where the two shapes were presented simultaneouslyon a black background in fixed positions on the screen(at the animal’s eye level, one appearing left of themiddle, and the other right, Fig. 1). Each trial wascomposed of one positive stimulus (S+) and one nega-tive stimulus (S−), which were positioned randomlyfrom trial to trial (left/right). Each session consisted of30 trials. When the positive stimulus was selected, bothstimuli disappeared, a short tone was emitted by thecomputer, and a food reward was provided. If the wrongstimulus was touched (S−), both stimuli disappeared, ashort buzz sounded, and a red screen was presented for3 s. In this case, a correction trial was immediatelyinitiated: the stimuli of the previous trial were presentedagain in the same positions. A correct choice terminatedthe trial and resulted in reward and presentation of a newtrial. After each trial (except correction trials), an inter-trial interval of 2 s was initiated (an empty black back-ground was presented). The learning criterion was set at≥20 correct first choices in 30 trials (66.7 %) in four outof five consecutive sessions. At this early stage in thetraining, the experimenter often needed to givedogs extra help in sessions, for example, verbalencouragement to approach the screen and touch,and occasional pointing. Therefore, the resultsfrom this test are presented only in the supplemen-tary materials (Table S1).

Touchscreen testing

Task 1: underwater photos and drawings discrimination

Once the criterion for the geometric form task wasreached, the dogs were transferred to a second discrim-ination training, involving three underwater photo-graphs, which had to be distinguished from three draw-ings (two of which were taken from posters byToulouse-Lautrec; Fig. 2b). The dogs were assigned totwo groups balanced for age group and sex. Group‘drawing’ was rewarded for touching the drawing andgroup ‘underwater’ was rewarded for touching the un-derwater photograph. In each trial, one of the three S+was randomly coupled side by side with one of the threeS−. The procedure and learning criterion were the sameas for the geometric form discrimination.

Task 2: clip art picture discrimination (training for task3: inferential reasoning by exclusion)

Once the dogs had completed the underwater photos anddrawing discrimination, they began the training for theinference by exclusion tests. Dogs were again split intotwo groups (Group ‘A’ and Group ‘B’) balanced for agegroup and sex. The dogs were trained to discriminatefour S+ and four S− stimuli (Fig. 3a), this time presentedon a white background. Once again, the forced two-choice procedure was utilised. The stimuli werecoloured clip art pictures obtained from the internetand were grouped within the two sets by avoidingsimilarities in colour, form or function. The clip artstimuli were the same as those used by Aust et al. inthe 2008 study. Each session consisted of 32 trials andcontained each of the 16 possible S+/S− pairings twiceper session. All dogs were required to reach two learn-ing criteria: a first learning criterion of ≥28 correct firstchoices (87.5 %) in two consecutive sessions and a finallearning criterion of ≥28 correct first choices in five ofseven consecutive sessions before beginning testing.Thirteen dogs which were tested prior to 2010 weretrained on a 100 % reward ratio. For the remaining 72dogs, the reward ratio was reduced stepwise to 75% (forexplanations of the rationale for a change in methodol-ogy, please see Supplementary Material: Reward ratioreduction). The unrewarded trials in the training servedto familiarise the dogs with the testing procedure, whichincluded up to eight unrewarded test trials in each ses-sion. Initially, training sessions for these dogs included

AGE (2016) 38: 6 Page 5 of 18 6

Fig. 1 a Schematic drawing of the apparatus and b photograph of a dog working in the testing niche with one side open

6 Page 6 of 18 AGE (2016) 38: 6

four trials that were not rewarded: i.e., the first choice ofany of the two stimuli terminated the trial without anyacoustic or visual feedback, correction trial or reward.The first learning criterion was utilised (≥28 correct firstchoices in two consecutive sessions), and once dogsreached this criterion, the reward ratio was further re-duced to six unrewarded trials per session. The samelearning criterion was applied again, after which a finaltraining phase with a 75% reward ratio (eight unreward-ed trials) was applied. The final learning criterion wasused for this phase (≥28 correct first choices in five ofseven consecutive sessions), the same criterion as wasused for the 13 dogs originally tested with the 100 %reward ratio.

Task 3: inferential reasoning by exclusion

Test 1: Test sessions consisted of 28 training trialswith four randomly interspersed test trials (a total of32 trials per session). The test trials contained fourknown S− from the training trials, which werepaired with four novel stimuli (Fig. 3b). The newstimuli (S’) replaced the S+ from the training. Eachof the 16 test combinations were shown twice, oncein cycle 1 (sessions 1–4) and once in cycle 2 (ses-sions 5–8). Subjects which choose by exclusionshould choose S’ due to inference of positive classmembership; i.e., by assuming, there is always amember of the positive class and by excluding S−due to its formed association with the negativeclass. But dogs which choose according to novelty(neophilia) or avoidance of S− should also chooseS’. In contrast, subjects which choose by familiarityshould prefer S−. Dogs which chose S’ in ≥22 outof a total of 32 test trials proceeded directly to test 2.

Test 2: In order to confirm that dogs chose byexclusion, an additional test was run to exclude thatdogs chose based on novelty or avoidance of S−.The subjects were again tested with one of the fourS’ paired with a known S− (same as test 1, Fig. 3b,hereafter known as the test 1 refresher) to refreshtheir memory, and then in one of the next two tothree trials, they were presented with the same S’paired with one of four novel alternative stimuli S^(Fig. 3c). If dogs chose by inference by exclusion,they would choose S’ when paired with the knownnegative (in tests 1 and 2 (in the test 1 refresher))and also choose S’ when S’ was paired with thenovel S^. Subjects which showed a preference forS’ in test 1 due to neophilia would now prefer themore novel S^ over S’ (novelty preference).Subjects which avoided S− in test 1 without mak-ing any inferences about the positive association ofS’ would choose randomly in test 2, showing nopreferences.

In each session in test 2, there were eight non-rewarded trials (four test 1 refresher and four test 2trials) interspersed within 24 training trials (32 trialsin total per session). Each of the 16 test combinations(four known S’ from test 1, paired with four novelstimuli (S^)) were again shown twice, once in cycle1 (sessions 1–4) and once in cycle 2 (sessions 5–8).

For each test 2 trial, dogs were scored as choosingby inference by exclusion if they firstly chose S’when paired with the known negative (test 1 refresh-er) and also chose S’ in the subsequent trial when S’was paired with the novel S^ (test 2 trial). Over theentire test 2, dogs were scored as choosing by infer-ence by exclusion above chance if they chose byexclusion in 13 or more out of the possible 32 test

Fig. 2 Training stimuli for the a geometric form and b underwater photo and drawing discriminations

AGE (2016) 38: 6 Page 7 of 18 6

trials (binomial test, chance level=0.25, p=0.016(chance level reflects the four possible choice com-binations of test 1 refresher, and test 2 trial; S’ andS’, S’ and S^, S− and S’, and finally S− and S^)).

Task 4: memory test

After completing the tests, all dogs had a minimum of6 months break before they were invited back to partic-ipate in a memory test consisting of a repetition of task 2(clip art picture discrimination/inference by exclusiontraining), up to the final criterion of ≥28 correct firstchoices (87.5 %) in five of seven consecutive sessions(Fig. 3d). Dogs, which had been trained on the 75 %reward ratio repeated the task at the 75 % reward ratio,and dogs, which were trained on the 100 % reward ratio,repeated the task at the 100 % reward ratio. The totalnumber of correct choices in the first session of thememory test was used as a measure of memory ability.

Data analysis

Statistical analyses were performed in R-3.0.1 (R CoreTeam 2013). Separate statistical models were calculatedfirst with age as a continuous variable (we tested forlinear and quadratic relationships) and then with age as acategorical variable to look for specific differences be-tween age groups. Results are presented as mean± standard deviation unless otherwise indicated.

In the geometric form, underwater photo and drawingdiscrimination and the clip art picture discrimination, weused the total number of sessions needed to reach crite-rion minus the minimum number of sessions needed toreach the criterion of each discrimination (in order tofulfill the assumptions for Poisson distribution) and thetotal number of correction trials as measures of learningspeed and behavioural flexibility. In the clip art picturediscrimination, the number of sessions needed to reachthe first criterion of ≥28 correct first choices in twoconsecutive sessions in both the 100 % rewarded andthe reduced reward groups was used to allow learningspeed to be assessed for the different reward ratios. Theproportion of test trial choices of S’ in test 1 and theproportion of test trials where dogs chose based oninference by exclusion (in the repetition of S’ pairedwith S− and the new S^ paired with S’) in test 2 werecalculated as two separate variables to describe thelogical reasoning strategies of the dogs. Finally, the totalnumber of correct choices in the first session of thememory test was used as a measure of memory ability.

Data were analysed using generalised linear modelsand generalised linear mixed models, with age, stimulusgroup, sex and neuter status included as fixed effects. Inthe inference by exclusion training and test 1, we alsoexamined the effect of the type of reward ratio (100 %reward or reduced reward). We included the two-wayinteraction between stimulus group and age to testwhether age effects differed between stimulus groups.When examining the proportion of test trial choices of

Fig. 3 a Reason by exclusiontraining stimuli, b test 1 stimuli, ctest 2 stimuli, and d memory teststimuli

6 Page 8 of 18 AGE (2016) 38: 6

S’ in test 1 and proportion of test trials where dogs chosebased on inference by exclusion in test 2, we alsochecked whether the dogs’ performance changed fromcycle 1 to cycle 2. The full models can be found in theSupplementary Materials (geometric forms discrimina-tion (Table S1), underwater photos and drawings dis-crimination (Table S2), clip art picture discrimination(Table S3), inferential reasoning by exclusion Test 1(Table S4), inferential reasoning by exclusion Test 2(Table S5), and memory test (Table S6)). Non-significant predictors (p>0.05) were then removed fromthe models and are not reported in the BResults^ section.According to the distribution of the response variables,models with negative binomial error structure and loglink function (Venables and Ripley 2002) were used forthe number of sessions to criterion and the total numberof correction trials, as well as models with binomialerror structure and logit link function for the proportionof choices of S’ in test 1 and test 2 and the proportion ofcorrect first choices in the memory test. When analysingdata including multiple data points per subject, dogidentity was included as a random factor in the model.Plots of residuals and Cook’s distance were examinedfor outliers. Since none of the data points exceededCook’s distance of 1, no outliers needed to be excluded.

Results

Task 1: underwater photo and drawing discrimination

Of the 95 dogs which began testing with the geometricform discrimination, 93 passed the learning criterion forthe underwater photos and drawing discrimination with-in 35 sessions. The number of sessions to criterionincreased linearly with age in months (Table 2: model1, Fig. 4a). The subsequent age group analysis revealedthat age groups 4 and 5 took significantly more sessionsto reach criterion compared to age group 1 (model 2).Dogs in the drawing group completed the task insignificantly fewer sessions than dogs in the un-derwater group, reflecting a difference in task dif-ficulty (Fig. 4a).

The total number of correction trials also increasedlinearly with age in months (Table 2: model 3, Fig. 4b).Age group 5 needed significantly more correction trialscompared to age group 1 (model 4). Dogs in the under-water group had significantly more correction trials than

dogs in the drawing group, furthermore supporting thedifference in task difficulty (Fig. 4b).

Task 2: clip art picture discrimination (training for task3: inferential reasoning by exclusion)

Of the 90 dogswhich began the training, 85 passed the firstlearning criterion of 28 or more correct choices in twoconsecutive sessions within 7 to 113 sessions. The fivedogs (all in age groups 4 and 5), which did not reach thelearning criterion, dropped out of the study due to motiva-tion problems. The number of sessions to criterion in-creased linearly with age in months (Table 3: model 5,Fig. 5a). Age groups 4 and 5 took significantly moresessions to reach criterion compared to age group 1 (model6). Dogs in group A completed the task in significantlyfewer sessions than dogs in groupB, reflecting a differencein task difficulty depending on the set of pictures the dogswere rewarded for (Table 3: model 5, Fig. 5a). Male dogsneeded more sessions to reach criterion than female dogs(males, 29.03±22.70, N=31: females, 23.48±16.26,N=54; Table 3: model 5). For further results and a discus-sion of these sex differences, please see SupplementaryMaterials. Dogs which participated in the reduced rewardratio training took significantly longer to reach the firstlearning criterion than dogs in the 100 % rewarded group(reduced reward 26.79±18.85, N=72: 100 % rewarded18.38±18.42, N=13; Table 3: model 5). Please refer toSupplementary Materials for additional results and a dis-cussion of the reward ratio reduction.

The total number of correction trials increased line-arly with age in months (Table 3: model 7, Fig. 5b). Agegroups 4 and 5 had significantly more correction trialscompared to age group 1 (model 8). Dogs in group Bhad significantly more correction trials than dogs ingroup A (Table 3: model 7, Fig. 5b). Male dogs neededmore correction trials than female dogs (males=217.26±159.46, females=198.52±200.80; Table 3: model 7).

Task 3: inferential reasoning by exclusion

Test 1: Of the 85 dogs which passed the first learningcriterion (≥28 correct first choices (87.5 %) in twoconsecutive sessions), 82 passed the final learningcriterion of 28 or more correct choices in five out ofseven consecutive sessions and participated in test 1.

The proportion of test trials in which dogs chose S’showed a significant increase with age in months(Table 4, Fig. 6). No significant differences between

AGE (2016) 38: 6 Page 9 of 18 6

the age groupswere detected, however. Dogs in groupB chose S’ in significantlymore test trials than dogs ingroup A (Table 4, Fig. 6a). Male dogs showed atendency to choose S’ more often than fe-males (males, N = 30, 0.69 ± 0.02, females,N= 52, 0.65 ± 0.01; Table 4). Dogs chose S’more often in cycle 1 compared to cycle 2 (Table 4,Fig. 6b). When results from cycles 1 and 2 werepooled, 42 (51 %) dogs preferred S’ (choose S’ in

22 or more test trials out of a total of 32) and thuschose based on exclusion (rejection of S− due to itsassociation with the negative class), novelty (selec-tion of S’ due to neophilia) or avoidance of theknown negative stimulus (S−) and proceeded totest 2 (apart from one dog which left the study atthis stage). The remaining dogs chose at chancelevel, apart from one individual, which chose basedon familiarity.

Table 2 Negative binomial generalised linear models showing the direction of effects and the significance level of the terms in theunderwater photos and drawings discrimination

Response variable Model Minimal model Average effect SE Wald statistic z p value

Number of sessionsto criterion

Model 1 Stimulus group: underwater 1.3841 0.1389 68.704 <0.001

Age in months 0.0072 0.0018 14.224 <0.001

Model 2 Age group 14.627 0.006

Age group 2 0.0109 0.1969 0.055 0.956

Age group 3 0.1200 0.2025 0.593 0.553

Age group 4 0.4832 0.1937 2.495 0.013

Age group 5 0.6104 0.2121 2.877 0.004

Number of correctiontrials

Model 3 Stimulus group: Underwater 1.7887 0.1470 88.076 <0.001

Age in months 0.0067 0.0022 9.584 0.002

Model 4 Age group 11.181 0.025

Age group 2 −0.0631 0.2135 −0.295 0.768

Age group 3 0.3723 0.2155 1.728 0.084

Age group 4 0.4144 0.2151 1.927 0.054

Age group 5 0.5741 0.2412 2.383 0.017

Z tests indicate which age groups differ from age group 1 in the respective analysis. Bold numbers indicate significant values at p= ≤0.05

Fig. 4 Line graph showing the linear relationship between age inmonths and a number of sessions to criterion and b number ofcorrection trials, shown separately for dogs that were rewarded for

choosing the underwater pictures and for dogs rewarded for choos-ing the drawings (with 95 % confidence intervals (dotted lines))

6 Page 10 of 18 AGE (2016) 38: 6

Test 2: There was no significant difference betweenthe number of times the dogs chose based on infer-ence by exclusion in cycle 1 and cycle 2, so datawere pooled and generalised linear models wereapplied (see Supplementary Material Table S5:model 11). Seven individuals (17 %) scored abovechance, and six of these seven were in group B(Fig. 7). The proportion of test trials in which the

dogs chose based on inference by exclusionshowed a significant increase with age in months(Table 5: model 12, Fig. 7). Age groups 3, 4 and 5chose S’ significantly more often compared to agegroup 1 (model 13). Dogs in group B chose byinference by exclusion in significantly more testtrials than dogs in group A (Table 5: model 12,Fig. 7).

Table 3 Negative binomial generalised linear models showing the direction of effects and the significance level of the terms in the clip artpicture discrimination (training for task 3: inferential reasoning by exclusion)

Response variable Model Minimal model Average effect SE Wald statistic z p value

Number of sessionsto criterion

Model 5 Age in months 0.0100 0.0017 32.326 <0.001

Stimulus group: B 0.2707 0.1095 5.908 0.015

Sex: male 0.3507 0.1169 8.710 0.003

Reward ratio 90 % 0.3486 0.1545 4.877 0.027

Model 6 Age group 29.633 <0.001

Age group 2 0.0612 0.2046 0.2990 0.765

Age group 3 0.1162 0.2088 0.5570 0.578

Age group 4 0.6525 0.2193 2.9750 0.003

Age group 5 0.8879 0.2215 4.0090 <0.001

Number of correctiontrials

Model 7 Age in months 0.0118 0.0019 37.953 <0.001

Stimulus group: B 0.4313 0.1250 11.169 <0.001

Sex: male 0.3184 0.1253 6.296 0.012

Model 8 Age group 32.130 <0.001

Age group 2 0.3174 0.2287 1.388 0.165

Age group 3 0.2992 0.2338 1.280 0.201

Age group 4 0.6798 0.2490 2.730 0.006

Age group 5 1.2756 0.2525 5.053 <0.001

Z tests indicate which age groups differ from age group 1 in the respective analysis. Bold numbers indicate significant values at p= ≤0.05

Fig. 5 Line graph showing the linear relationship between age in months and a number of sessions to criterion and b number of correctiontrials, separately for groups A and B (with 95 % confidence intervals (dotted lines))

AGE (2016) 38: 6 Page 11 of 18 6

The proportion of test trials in which dogs choseby exclusion showed a significant increase with thetotal number of correction trials in the inference byexclusion training (Table 5, model 15) after con-trolling for age in months. Therefore, regardless ofage, dogs which needed more correction trials inthe training chose more often using inference byexclusion in test 2.

Task 4: memory test

Of the 82 dogs which completed the final learningcriterion of the inference training, 46 participated inthe memory test after a break of at least 6 months.Forty-two of these dogs scored significantly abovechance level in the first session (22 or more out of thepossible 32 first correct choices (binomial test 22/32 = 0.6875, chance level = 0.5, p = 0.050; 81.52±10.10 %). There were no significant effects of age or

stimulus group on the proportion of correct first choicesin the first session of the memory test (SupplementaryTable S6).

Discussion

The aim of the present study was to examine age effectson visual discrimination learning, inferential reasoningby exclusion and long-term memory in domestic dogskept as pets. We found a significant effect of age on thenumber of trials needed to reach criterion (as age in-creased, discrimination learning ability decreased) anddegree of perseveration (the number of correction trials)in the two visual discrimination learning tasks. In con-trast, older dogs chose more often by exclusion thanyounger dogs in the crucial (second) reasoning by ex-clusion test. Finally, dogs’ long-termmemorywas main-tained into old age, with no difference in performance in

Table 4 Generalised linear mixed model on the proportion of trials chose S’ when paired with a known negative (S−) in test 1 of theinference by exclusion task, showing the direction of effects and the significance level of the terms

Response variable Model Minimal model Average effect SE Wald statistic /deviance p value

Proportion of trials chose S’ Model 9 Cycle: cycle 2 −0.4943 0.0839 34.723 <0.001

Stimulus: group B 0.3478 0.1007 11.136 <0.001

Age in months 0.0037 0.0014 6.567 0.010

Sex: male 0.1919 0.0988 3.693 0.055

Bold numbers indicate significant values at p =≤0.05

Fig. 6 The proportion of test trials in test 1 in which the dog choseS’; a group A and group B, and b cycle 1 (sessions 1 to 4) andcycle 2 (sessions 5 to 8), and age in months. The upper dashed lineindicates the levels of performance beyondwhich preference for S’

was inferred (68.75 %; choice by novelty, avoidance of S− orreasoning by exclusion). The lower dashed line indicates the levelof performance below which preference for S− was inferred(31.25 %; choice by familiarity)

6 Page 12 of 18 AGE (2016) 38: 6

any of the age groups after a 6-month break from thetouchscreen.

The ability to learn new visual stimulus associationsdecreased with age as predicted. The youngest dogsaged from 5 months to 1 year needed the lowest numberof sessions to complete the criteria, indicating that thisage group was already performing at peak performance,and from this age onward, dogs’ learning abilities beganto decline. In contrast to the present study, previousstudies in non-human animals have found no effect of

aging on associative learning in simple object discrim-ination tasks either in the rhesus macaque (aged from 3to 34 years; Bachevalier et al. 1991) or in laboratorydogs (aged from 1.5 to 11 years; Milgram et al. 1994).One possible reason for this discrepancy is that, byutilising a higher number of stimuli to be discriminated,we sufficiently increased the difficulty level and thusfacilitated the appearance of age effects. This interpre-tation is also supported by the difference we find be-tween the two stimuli groups both in the drawing andunderwater photo discrimination and in the clip artdiscrimination: If the discrimination seems to be easierfor the dogs (‘drawing’; group B), the age differences,although still apparent, are not as pronounced as in themore difficult groups (‘underwater’; group A).However, although age effects were more apparent inthe groups with the less preferred stimuli as positive(that is, in the more difficult version of each task), wefound no evidence for an interaction between age andstimulus group in any of the discrimination tasks. For adiscussion of stimulus preferences in two choice dis-criminations, please refer to the SupplementaryMaterials: Stimulus preferences.

Age differences were more pronounced in the clip artpicture discrimination than in the drawing and underwa-ter photo discrimination. This difference in effect sizemay be explained firstly in terms of the number ofstimuli to be discriminated (six in the drawing andunderwater discrimination and eight in the picture dis-crimination) and additionally by the fact that the draw-ing discrimination could be solved more easily by

Fig. 7 The proportion of times in which the dog chose based oninference by exclusion in group A and group B and age in monthsin test 2 (cycles 1 and 2 pooled). The dashed line indicates thelevels of performance beyond which preference for S’was inferred(40.625 %; reasoning by exclusion)

Table 5 Generalised linear model on the proportion of times thedogs’ chose S’ when paired with the known negative (test 1refresher) and also chose S’ in the subsequent trial when S’ was

paired with the novel S^ (test 2 trial) in the inference by exclusiontask, showing the direction of effects and the significance level ofthe terms

Response variable Model Minimal model Averageeffect

SE Wald statistic /deviance

z p value

Proportion of times chose S’ inboth test 1 refresher trialand test 2 trial

Model 12 Age in months 0.0099 0.0014 45.538 <0.001

Stimulus: group B 0.7027 0.1367 27.739 <0.001

Model 13 Age group 54.570 <0.001

Age group 2 0.4654 0.2816 1.653 0.094

Age group 3 0.6387 0.2989 2.137 0.033

Age group 4 1.2223 0.2900 4.215 <0.001

Age group 5 1.3916 0.2788 4.992 <0.001

Model 14 Sessions to criterion 0.0008 0.0029 0.082 0.775

Model 15 Total no. of correction trials 0.0006 0.0003 4.103 0.043

Z tests indicate which age groups differ from age group 1 in the respective analysis. Age in months was included in models 12 and 13 tocontrol for age effects. Bold numbers indicate significant values at p=≤0.05

AGE (2016) 38: 6 Page 13 of 18 6

learning a perceptual discrimination rule. All the draw-ings looked perceptually similar to each other, as did theunderwater photographs, but the clip art picture discrim-ination required that all the stimuli be encoded intomemory individually, as there were no perceptual com-monalities in the positive or the negative stimuli. Ourresults are in line with the findings from human studies;age effects can be better detected bymore complex tasks(Alvarez and Emory 2006; Mell et al. 2005).

The poorer performance of dogs aged over 3 years inour study could be explained by several possibilities. First,older dogs may suffer from attentional deficits due toreduced processing resources (Snigdha et al. 2012).Additionally, older dogs may use ineffectual strategies inan attempt to solve the discriminations, for example, astimulus response strategy (such as stimulus preferencesor avoidance, as seenwhen dogs repeatedlymake incorrectchoices) and/or a positional strategy (side bias), beforefinally switching to a cognitive strategy. Both stimulusresponse and positional strategies require less workingmemory and are therefore less costly than a cognitivestrategy (Chan et al. 2002). Unfortunately, we were unableto analyse positional strategies due to limitations in thesoftware program.

Second, younger dogs may have been quicker to utilisethe cognitive strategy of forming reward associations forthe positive stimuli by utilising working memory andswift-encoding to long-termmemory. These younger dogs,assuming that their working memory abilities were good,might have shown more focused selective attentionallowing them to quickly pick out the correct stimuli andignore the negative stimuli (Mongillo et al. 2010; Snigdhaet al. 2012;Wallis et al. 2014). In contrast, older dogs havea reduced capacity for working memory (Chan et al. 2002;Tapp et al. 2003b), similarly to other species includinghumans (Cowan 2001; Matzel and Kolata 2010).Evidence in humans suggests that older individuals withlower working memory capacity may also need to copewith the processing of negative (or distractor) stimuli,which leads to slower learning and the storage of moreinformation inmemory than younger individuals with highworking memory capacity (Konstantinou et al. 2014;Vogel et al. 2005).

Third, an important non-cognitive factor, whichcould have influenced the results, is age differences insensory ability (namely eyesight). However, all olderdogs in our study were able to pass the criteria in threevisual discrimination tasks, and in the geometric formstask, we found no age differences in the number of

sessions to criteria (see Supplementary Materials,Table S1). Additionally, we tested many of the subjectsin behavioural tests and found little evidence that visualimpairments influenced the dogs’ performance (Walliset al. 2015; Wallis et al. 2014).

The total number of correction trials increased withage in all discrimination tasks possibly due to a lack ofattention, persistency and/or side bias in the older dogs,resulting in an inability to adjust thinking or attention inresponse to feedback. Similarly to earlier findings indogs (Chan et al. 2002), the oldest age group displayedthe most perseverative errors and thus displayed reducedflexibility. Aged members of other species have alsoshown reduced flexibility reflected in an inability tosuppress and/or change behaviour on the basis of nega-tive feedback; for example rats (Stephens et al. 1985),non-human primates (Lai et al. 1995; Manrique and Call2015; Voytko 1999; Voytko 1993) and humans(Botwinick 1978; Daigneault et al. 1992).

The proportion of test trials in which the dogs chosebased on novelty, avoidance or exclusion in test 1 of theinference by exclusion task increased with age.However, no significant differences between the agegroups were found. The proportion of test trials in whichthe dogs chose based on exclusion in test 2 also in-creased with age, but with most dogs choosing at chancelevels. Less than 10 % of dogs in the current studyshowed patterns of choice consistent with inference byexclusion, indicating that inference by exclusion wasnot the predominant strategy used by the dogs. In Austet al.’s (2008) study by comparison, three out of six dogswere found to display this ability.

In contrast to our prediction of a peak in inference byexclusion ability in young adult dogs, seven dogs inmiddle-to-late adulthood were found to perform abovechance, suggesting that they used reasoning by exclu-sion. Similarly, in non-human primates, one study byCall (2006) found that the ability to reason by exclusionincreases with age. Our results are superficially similarto the primate study; however, after looking into the datamore carefully, our results seem to reflect a learningrather than a reasoning effect. This learning effect wasstrongest in younger individuals: In the test trials, thedogs were not rewarded for choosing based on exclu-sion (choosing S’), which might have made them switchto choosing randomly due to the missing feedback.

A similar effect might explain why in test 1 choosing S’(based on novelty, avoidance or exclusion) declined fromthe first to the second cycle. In the tests, younger dogs

6 Page 14 of 18 AGE (2016) 38: 6

might have reacted to the lack of feedback sooner/moreoften than the older dogs, reflecting their more flexibleproblem solving style. This interpretation is further sup-ported by the impact of the degree of perseverativeresponding in the training on performance in the inferenceby exclusion in test 2. After controlling for age, our resultsindicated that a higher amount of perseverative respondingincreases the likelihood of finding response patterns con-sistent with choosing by exclusion. Conversely, the higherdegree of flexibility of the younger dogs may have led to alower probability of choices following the inference byexclusion pattern in this particular paradigm, where testtrials were not rewarded. We suggest that older dogs,especially those that were in the more difficult to learngroup B, were more likely to stick with their initial choiceof S’ due to the fact that they showed greater levels ofperseverative responding in the training and consequentlyhadmore chance to learn about the negative stimuli. Thesedogs may have persisted in their choice of S’ in the testtrials in test 1, did not alter their strategy in response to thelack of feedback, and may have been able to encode S’ toworkingmemory to enable them to choose S’when pairedwith S^ a few trials later in test 2. In the study of Aust et al.(2008), all three dogs, which chose by inference by exclu-sion, and which were also in group B, needed moresessions to reach criteria in the training and therefore hadmore experience with correction trials, similarly to dogs inour study. Results from studies on aged humans showsimilar findings of reduced flexibility (shown in difficultiesin switching task sets) and deficiencies in adaptation toexternal feedback (Kray and Lindenberger 2000; Mellet al. 2005), supporting the findings of the current study.

Finally, there was no effect of age or stimulus group onthe performance of dogs in the memory test 6 months later.However, the 6-month break was likely too short a timeperiod to enable the detection of age effects. The lack of ageeffects on long-term memory confirms previous results inlaboratory dogs by Araujo et al. (2005). Nearly all the dogstested in the current study scored above chance in the veryfirst session, suggesting that long-termmemory for specificstimuli on the touchscreen is longer than 6 months in dogs.Recently, we re-tested five dogs of different breeds, whichhad undergone inference by exclusion training between 3and 5 years previously, and these individuals performed atover 80% correct first choices on the first day of re-training,which is comparable to the performance of dogs in thememory test of the current study. Therefore, domestic dogs’long-term memory for picture stimuli may exceed 5 years,similarly to baboons and pigeons (Fagot and Cook 2006).

In conclusion, older dogs showed slower learning andreduced flexibility, which may have contributed to anincrease in choosing by inference by exclusion in the testsin comparison to young dogs, which were more sensitiveto the lack of feedback in test trials, and subsequentlyflexibly changed their response pattern and used strategiesother than inference by exclusion. Dogs’ long-term mem-ory for the clip art picture discrimination was well main-tained into old age. Our results in the visual discriminationlearning tasks show clear age differences confirming thatthe tests used are suitable to detect cognitive aging in petdogs and provide additional evidence of the suitability ofthe dog as a model for aging. The baseline measuresassociated with normal cognitive aging in the pet BorderCollie found in the current study can serve as a basis forcomparison to help diagnose cognition-related problemsand as a tool to assist with the development of treatmentsto delay cognitive decline. Moreover, the touchscreenapparatus offers a standardised procedure, which can beapplied across different dog breeds, other non-humananimals and even humans. Utilising this method, futurestudies could investigate the development and aging ofcognitive processes and disorders and their interactionswith genetic, environmental and social factors.

Acknowledgments We would like to thank the owners whovolunteered to participate in this long-term study and especiallythe research assistants Angela Gaigg and Teresa Marmota fortraining and testing the dogs on the touchscreen. Additionally,we would like to thank Mark O’Hara for statistical help and oursponsors Royal Canin for providing funding for this project. LisaWallis was furthermore supported by the DK CogCom Program(Austrian Science Fund Doctoral Programs W1234). Writing wassupported by a FWF grant (project number P24840-B16) to FR,WWTF project CS11-026 to ZsV, WWTF project CS11-025 toLH, and the FWF grant P21418 to LH and FR.

Open Access This article is distributed under the terms of theCreative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestrict-ed use, distribution, and reproduction in any medium, providedyou give appropriate credit to the original author(s) and the source,provide a link to the Creative Commons license, and indicate ifchanges were made.

References

Adams B, Chan A, Callahan H, Milgram NW (2000a) The canineas a model of human cognitive aging: recent developments.Prog Neuro-Psychopharmacol Biol Psychiatry 24:675–692

AGE (2016) 38: 6 Page 15 of 18 6

Adams B, Chan A, Callahan H, Siwak C, Tapp D, Ikeda-DouglasC,…, Milgram NW (2000). Use of a delayed non-matchingto position task to model age-dependent cognitive decline inthe dog. Behav Brain Res 108:47–56. doi:10.1016/S0166-4328(99)00132-1

Alvarez JA, Emory E (2006) Executive function and the frontallobes: a meta-analytic review. Neuropsychol Rev 16(1):17–42. doi:10.1007/s11065-006-9002-x

Araujo JA, Studzinski CM,Milgram NW (2005) Further evidencefor the cholinergic hypothesis of aging and dementia from thecanine model of aging. Prog Neuro-Psychopharmacol BiolPsychiatry. doi:10.1016/j.pnpbp.2004.12.008

Aust U, Range F, Steurer M, Huber L (2008) Inferential reasoningby exclusion in pigeons, dogs, and humans. Anim Cogn11(4):587–597. doi:10.1007/s10071-008-0149-0

Bachevalier J, Landis LS,Walker LC, BricksonM,MishkinM, PriceDL, Cork LC (1991) Aged monkeys exhibit behavioral deficitsindicative of widespread cerebral dysfunction. Neurobiol Aging12(2):99–111. doi:10.1016/0197-4580(91)90048-O

Baddeley AD, Baddeley HA, Bucks RS, Wilcock GK (2001)Attentional control in Alzheimer’s disease. Brain J Neurol124:1492–1508. doi:10.1093/brain/124.8.1492

Baltes PB (1987) Theoretical propositions of life-span develop-mental psychology: on the dynamics between growth anddecline. Dev Psychol 23(5):611–626. doi:10.1037//0012-1649.23.5.611

Borella E, Carretti B, De Beni R (2008) Working memory andinhibition across the adult life-span. Acta Psychol 128(1):33–44. doi:10.1016/j.actpsy.2007.09.008

Botwinick J (1978) Aging and behavior: a comprehensive inte-gration of research findings. Springer, New York

Brickman AM, Stern Y (2010) Aging and memory in humans. In:Squire L (ed), Encyclopedia of Neuroscience (Vol. 1, pp. 175–180). Elsevier. doi:10.1016/B978-008045046-9.00745-2

Brockmole JR, Logie RH (2013) Age-related change in visualworking memory: a study of 55,753 participants aged 8–75.Front Psychol 4:12. doi:10.3389/fpsyg.2013.00012

Bussey TJ, Padain TL, Skillings EA, Winters BD, Morton AJ,Saksida LM (2008) The touchscreen cognitive testing meth-od for rodents: how to get the best out of your rat. LearnMem15(7):516–523. doi:10.1101/lm.987808

Call J (2006) Inferences by exclusion in the great apes: the effect ofage and species. Anim Cogn 9(4):393–403. doi:10.1007/s10071-006-0037-4

Cepeda NJ, Kramer AF, Gonzalez de Sather JC (2001) Changes inexecutive control across the life span: examination of task-switching performance. Dev Psychol 37(5):715–730. doi:10.1037/0012-1649.37.5.715

Chan ADF, Nippak PMD, Murphey H, Ikeda-douglas CJ,Muggenburg B, Head E, …, Milgram NW (2002)Visuospatial impairments in aged canines (Canis familiaris):the role of cognitive-behavioral flexibility. Behav Neurosci116(3):443–454. doi:10.1037/0735-7044.116.3.443

Christie L-A, Studzinski CM, Araujo JA, Leung CSK, Ikeda-Douglas CJ, Head E,…, Milgram NW (2005) A comparisonof egocentric and allocentric age-dependent spatial learningin the beagle dog. Prog Neuro-Psychopharmacol BiolPsychiatry 29(3):361–369. doi:10.1016/j.pnpbp.2004.12.002

Chuderski A (2013) When are fluid intelligence and workingmemory isomorphic and when are they not? Intelligence41(4):244–262. doi:10.1016/j.intell.2013.04.003

Clark CR, Paul RH,Williams LM, ArnsM, Fallahpour K, HandmerC, Gordon E (2006) Standardized assessment of cognitivefunctioning during development and aging using an automatedtouchscreen battery. Arch Clin Neuropsychol Off J Nat AcadNeuropsychol 21(5):449–467. doi:10.1016/j.acn.2006.06.005

Costello MC, Madden DJ, Mitroff SR, Whiting WL (2010) Age-related decline of visual processing components in change de-tection. Psychol Aging 25(2):356–368. doi:10.1037/a0017625

Cowan N (2001) The magical number 4 in short-term memory: areconsideration of mental storage capacity. Behav Brain Sci24(1):87–114. doi:10.1017/S0140525X01003922,discussion 114–185

Craik FIM, Bialystok E (2006) Cognition through the lifespan:mechanisms of change. Trends Cogn Sci 10(3):131–138. doi:10.1016/j.tics.2006.01.007

Daigneault S, Braun CMJ, Whitaker HA (1992) Early effects ofnormal aging on perseverative and non-perseverative pre-frontal measures. Dev Neuropsychol 8(1):99–114. doi:10.1080/87565649209540518

De Luca CR, Wood SJ, Anderson V, Buchanan J, Proffitt TM,Mahony K, Pantelis C (2003) Normative data from theCANTAB. I: development of executive function over thelifespan. J Clin Exp Neuropsychol 25:242–254. doi:10.1076/jcen.25.2.242.13639

Fagot J, Cook RG (2006) Evidence for large long-term memorycapacities in baboons and pigeons and its implications forlearning and the evolution of cognition. Proc Natl Acad Sci US A 103(46):17564–17567. doi:10.1073/pnas.0605184103

Fiset S (2007) Landmark-based search memory in the domesticdog (Canis familiaris). J Comp Psychol 121(4):345–353. doi:10.1037/0735-7036.121.4.345

Fiset S, Beaulieu C, Landry F (2003) Duration of dogs’ (Canisfamiliaris) working memory in search for disappearing objects.Anim Cogn 6(1):1–10. doi:10.1007/s10071-002-0157-4

González-Martínez Á, Rosado B, Pesini P, García-Belenguer S,Palacio J, Villegas A, …, Sarasa M (2013) Effect of age andseverity of cognitive dysfunction on two simple tasks in petdogs. Vet J 198(1):176–181. doi:10.1016/j.tvjl.2013.07.004

Grant DA, Berg E (1948) A behavioral analysis of degree ofreinforcement and ease of shifting to new responses in aWeigl-type card-sorting problem. J Exp Psychol 38(4):404–411. doi:10.1037/h0059831

Griebel U, Oller DK (2012) Vocabulary learning in a Yorkshireterrier: slowmapping of spoken words. PLoS ONE 7(2). doi:10.1371/journal.pone.0030182

Head E, Mehta R, Hartley J, Kameka M, Cummings BJ, CotmanCW, …, Milgram NW (1995) Spatial learning and memoryas a function of age in the dog. Behav Neurosci 109(5):851–858. doi:10.1037/0735-7044.109.5.851

Head E, CallahanH,Muggenburg BA, Cotman CW,MilgramNW(1998) Visual-discrimination learning ability and beta-amyloid accumulation in the dog. Neurobiol Aging 19(5):415–425. doi:10.1016/S0197-4580(98)00084-0

Head E, Cotman CW, Milgram NW (2000) Canine cognition,aging and neuropathology. Prog Neuro-PsychopharmacolBiol Psychiatry 24(5):671–673. doi:10.1016/S0278-5846(00)00100-7

Heibeck TH, Markman EM (1987) Word learning in children: anexamination of fast mapping. Child Dev 58(4):1021–1034.doi:10.2307/1130543

6 Page 16 of 18 AGE (2016) 38: 6

Herman LM, Richards DG, Wolz JP (1984) Comprehension ofsentences by bottlenosed dolphins. Cognition 16(2):129–219. doi:10.1016/0010-0277(84)90003-9

Horst JS, Samuelson LK (2008) Fast mapping but poor retentionin 24-month-old infants. Infancy 13(2):128–157. doi:10.1080/15250000701795598

Huber L, Apfalter W, Steurer M, Prossinger H (2005) A newlearning paradigm elicits fast visual discrimination in pi-geons. J Exp Psychol Anim Behav Process 31(2):237–246.doi:10.1037/0097-7403.31.2.237

Joly M, Ammersdörfer S, Schmidtke D, Zimmermann E (2014)Touchscreen-based cognitive tasks reveal age-related impair-ment in a primate aging model, the grey mouse lemur(Microcebus murinus). PLoS ONE 9(10):e109393. doi:10.1371/journal.pone.0109393

Julesz B, Schumer RA (1981) Early visual perception. Annu RevPsychol 32:575–627. doi:10.1146/annurev.ps.32.020181.003043

Kaminski J, Call J, Fischer J (2004) Word learning in a domesticdog: evidence for Bfast mapping^. Science 304(5677):1682–1683. doi:10.1126/science.1097859

Kastak CR, Schusterman RJ (2002) Sea lions and equivalence:expanding classes by exclusion. J Exp Anal Behav 78(3):449–465. doi:10.1901/jeab.2002.78-449

Kaulfuss P, Mills DS (2008) Neophilia in domestic dogs (Canisfamiliaris) and its implication for studies of dog cognition.Anim Cogn 11(3):553–556. doi:10.1007/s10071-007-0128-x

Konstantinou N, Beal E, King JR, Lavie N (2014). Workingmemory load and distraction: dissociable effects of visualmaintenance and cognitive control. Atten PerceptPsychophys 1985–1997. doi:10.3758/s13414-014-0742-z

Kray J, Lindenberger U (2000) Adult age differences in taskswitching. Psychol Aging 15(1):126–147. doi:10.1037/0882-7974.15.1.126

Kyllonen PC, Christal RE (1990) Reasoning ability is (little morethan) working-memory capacity?! Intelligence. doi:10.1016/S0160-2896(05)80012-1

Lai ZC, Moss MB, Killiany RJ, Rosene DL, Herndon JG (1995)Executive system dysfunction in the aged monkey: spatialand object reversal learning. Neurobiol Aging 16(6):947–954. doi:10.1016/0197-4580(95)02014-4

Landsberg GM, HunthausenWL, Ackerman LJ (2003) The effects ofaging on behavior in senior pets. In: Saunders (ed) Behaviorproblems of the dog and cat, 2nd edn. Elsevier Health Sciences,Philadelphia, pp 269–280, Retrieved from https://books.google.com/books?id=eYbVBMkYvSAC&pgis=1

Lavie N (1995) Perceptual load as a necessary condition forselective attention. J Exp Psychol Hum Percept Perform21(3):451–468. doi:10.1037/0096-1523.21.3.451

Lee JY, Lyoo IK, Kim SU, Jang HS, Lee DW, Jeon HJ, …, ChoMJ (2005) Intellect declines in healthy elderly subjects andcerebellum. Psychiatry Clin Neurosci 59(1):45–51. doi:10.1111/j.1440-1819.2005.01330.x

Manrique HM, Call J (2015) Age-dependent cognitive inflexibil-ity in great apes. Anim Behav 102:1–6. doi:10.1016/j.anbehav.2015.01.002

Matzel L, Kolata S (2010) Selective attention, working memory,and animal intelligence. Neurosci Biobehav Rev 34(1):23–30. doi:10.1016/j.neubiorev.2009.07.002.Selective

Mell T, Heekeren HR, Marschner A, Wartenburger I, Villringer A,Reischies FM (2005) Effect of aging on stimulus-reward

association learning. Neuropsychologia 43(4):554–563. doi:10.1016/j.neuropsychologia.2004.07.010

Milgram NW, Head E, Weiner E, Thomas E (1994) Cognitivefunctions and aging in the dog: acquisition of nonspatialvisual tasks. Behav Neurosci 108(1):57–68. doi:10.1037/0735-7044.108.1.57

Milgram NW, Head E, Muggenburg B, Holowachuk D, MurpheyH, Estrada J, …, Cotman CW (2002) Landmark discrimina-tion learning in the dog: effects of age, an antioxidant forti-fied food, and cognitive strategy. Neurosci Biobehav Revdoi:10.1016/S0149-7634(02)00039-8

Milgram NW, Head E, Zicker SC, Ikeda-Douglas CJ, Murphey H,Muggenburg B, …, Cotman CW (2005) Learning ability inaged beagle dogs is preserved by behavioral enrichment anddietary fortification: a two-year longitudinal study. NeurobiolAging 26(1):77–90. doi:10.1016/j.neurobiolaging.2004.02.014

Mongillo P, Bono G, Regolin L, Marinelli L (2010) Selectiveattention to humans in companion dogs, Canis familiaris.Anim Behav 80(6):1057–1063. doi:10.1016/j.anbehav.2010.09.014

Mongillo P, Araujo JA, Pitteri E, Carnier P, Adamelli S, Regolin L,Marinelli L (2013) Spatial reversal learning is impaired byage in pet dogs. Age 35:2273–2282. doi:10.1007/s11357-013-9524-0

Moshman D (2004) From inference to reasoning: the constructiono f r a t i o n a l i t y. T h i n k Re a s on . d o i : 1 0 . 1 0 8 0 /13546780442000024

Nagahara AH, Bernot T, Tuszynski MH (2010) Age-related cog-nitive deficits in rhesus monkeys mirror human deficits on anautomated test battery. Neurobiol Aging 31(6):1020–1031.doi:10.1016/j.neurobiolaging.2008.07.007

Pearce JM (2008) Animal learning and cognition: an introduction,3rd edn. Psychology Press, New York

Pilley JW, Reid AK (2011) Border collie comprehends objectnames as verbal referents. Behav Process 86(2):184–195.doi:10.1016/j.beproc.2010.11.007

R. C. Team (2013) R: a language and environment for statisticalcomputing. R Foundation for Statistical Computing, Vienna,Retrieved from http://www.r-project.org

Range F, Aust U, Steurer M, Huber L (2008) Visual categorizationof natural stimuli by domestic dogs. Anim Cogn 11(2):339–347. doi:10.1007/s10071-007-0123-2

Rapp PR (1990) Visual discrimination and reversal learning in theaged monkey (Macaca mulatta). Behav Neurosci 104(6):876–884, Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/2285486

Salvin HE, McGreevy PD, Sachdev PS, Valenzuela MJ (2011)The canine sand maze: an appetitive spatial memory para-digm sensitive to age-related change in dogs. J Exp AnalBehav 95(1):109–118. doi:10.1901/jeab.2011.95-109

Sander MC, Lindenberger U, Werkle-Bergner M (2012) Lifespanage differences in workingmemory: a two-component frame-work. Neurosci Biobehav Rev 36(9):2007–2033. doi:10.1016/j.neubiorev.2012.06.004

Scott W (1962) Cognitive complexity and cognitive flexibility.Sociometry 25(4):405–414. doi:10.2307/2785779

Siegal M, Barlough JE (1995) UC Davis book of dogs. HarperCollins, New York

Snigdha S, Christie L-AA, De Rivera C, Araujo JA, MilgramNW,Cotman CW (2012) Age and distraction are determinants ofperformance on a novel visual search task in aged Beagle

AGE (2016) 38: 6 Page 17 of 18 6

dogs. Age (Dordr) 34(1):67–73. doi:10.1007/s11357-011-9219-3

Spiegel C, Halberda J (2011) Rapid fast-mapping abilities in 2-year-olds. J Exp Child Psychol 109(1):132–140. doi:10.1016/j.jecp.2010.10.013

Spinelli S, Pennanen L, Dettling AC, Feldon J, Higgins GA, PryceCR (2004) Performance of the marmoset monkey on com-puterized tasks of attention and working memory. CognBrain Res 19(2):123–137. doi:10.1016/j.cogbrainres.2003.11.007

Stephens DN, Weidmann R, Quartermain D, Sarter M (1985)Reversal learning in senescent rats. Behav Brain Res 17(3):193–202. doi:10.1016/0166-4328(85)90043-9

Steurer MM, Aust U, Huber L (2012) The Vienna comparativecognition technology (VCCT): an innovative operant condi-tioning system for various species and experimental proce-dures. Behav Res Methods 44(4):909–918. doi:10.3758/s13428-012-0198-9

Süß HM, Oberauer K, Wittmann WW, Wilhelm O, Schulze R(2002) Working-memory capacity explains reasoning abili-ty—and a little bit more. Intelligence 30:261–288. doi:10.1016/S0160-2896(01)00100-3

Tapp PD, Siwak CT, Estrada J, Head E,Muggenburg BA, CotmanCW, Milgram NW (2003a) Size and reversal learning in thebeagle dog as a measure of executive function and inhibitorycontrol in aging. Learn Mem 10(1):64–73. doi:10.1101/lm.54403

Tapp PD, Siwak CT, Estrada J, Holowachuk D, Milgram NW(2003b) Effects of age on measures of complex workingmemory span in the beagle dog (Canis familiaris) using twoversions of a spatial list learning paradigm. Learn Mem10(2):148–160. doi:10.1101/lm.56503

Venables WN, Ripley BD (2002) Modern applied statistics with S.Springer New York, New York. doi:10.1007/978-0-387-21706-2

Vogel EK, McCollough AW, Machizawa MG (2005) Neural mea-sures reveal individual differences in controlling access toworking memory. Nature 438(7067):500–503. doi:10.1038/nature04171

Voytko ML (1993) Cognitive changes during normal aging inmonkeys assessed with an automated test apparatus.Neurobiol Aging 14(6):643–644. doi:10.1016/0197-4580(93)90055-G

Voytko ML (1999) Impairments in acquisition and reversals oftwo-choice discriminations by aged rhesus monkeys.Neurobiol Aging 20(6):617–627. doi:10.1016/S0197-4580(99)00097-4

Wallis LJ, Range F, Müller CA, Serisier S, Huber L, Virányi Z(2014) Lifespan development of attentiveness in domesticdogs: drawing parallels with humans. Front Psychol 5(71):71. doi:10.3389/fpsyg.2014.00071

Wallis LJ, Range F, Müller CA, Serisier S, Huber L, Virányi Z(2015) Training for eye contact modulates gaze following indogs. Anim Behav 106:27–35. doi:10.1016/j.anbehav.2015.04.020

6 Page 18 of 18 AGE (2016) 38: 6